Environmental applications of gas sensor arrays:: combustion atmospheres and contaminated soils

Gas sensor arrays have been used to analyze volatile organic compounds in contaminated soils and exhaust gases coming from combustion processes. Sputtered thin films of semiconductor metal oxides were used as gas sensors in the sensor arrays. Combustion gases such as NOx, SO2, and benzene were detected in a highly toxic atmosphere formed by N-2, O-2, H2S, HF, HCl and water vapour. Sensitivities by 100% were obtained for different sensors when exposed to NOx and SO2. Six volatile organic compounds coming from contaminated soils were successfully identified using different pattern recognition methods such as principal component analysis and backpropagation neural networks. In both cases the use of the normalized fractional conductance change as preprocessing algorithm was decisive. Quantitative determinations of the mixtures of the volatile compounds were performed with relative prediction errors ranging from 2 to 50% for the calibration set. Higher errors were found using the validation data set. Backpropagation neural networks with partially connected hidden layer resulted in general more satisfactory than multiple linear regression methods because the response models were not well satisfied for all the sensors. The use of the normalized fractional conductance change as preprocessing algorithm gave the best results with the neural networks. (C) 1999 Elsevier Science S.A. All rights reserved.